Bayesian Understanding for Mediator Selection Framework

A collection of quantitative tools for selecting mediating effects within exploratory Bayesian mediation models. The package accommodates both continuous and dichotomous outcomes, including the dependent variables and the mediators for identifying and analyzing mediation pathways.

buzzMed v0.1.4

Bayesian Mediation Analysis with Variable Selection

Overview

buzzMed is an R package for Bayesian mediation analysis. It provides tools for exploratory Bayesian mediation models with Bayesian variable selection, supporting continuous and binary mediators and outcomes. The package also includes a longitudinal Bayesian mediation model for repeated-measures data.

Features

• Exploratory Bayesian mediation analysis with variable selection
• Continuous or binary (0/1) mediators
• Continuous or binary (0/1) outcomes
• Multiple predictors and candidate mediators
• Formula-based model specification
• Longitudinal Bayesian mediation for repeated-measures data

Requirements

This package requires JAGS (Just Another Gibbs Sampler) to be installed on your system.

Download JAGS from:

https://mcmc-jags.sourceforge.io/

Installation

Install the development version from GitHub:

# install.packages("remotes")
remotes::install_github("olfactorybulb/buzzMed")

library(buzzMed)

Main Functions

Generalized Two-stage(GT) exploratory Bayesian mediation model

The package provides four functions for exploratory Bayesian mediation analysis based on the mediator and outcome variable types.

Longitudinal Bayesian Mediation

• longBMed() fits a Bayesian mediation model for repeated-measures data with one or more predictors, multiple candidate mediators, and a continuous outcome.

Example Usage

GT-Exploratory Bayesian Mediation

library(buzzMed)

# Create toy data
my_data <- data.frame(
  MyPredictor = rnorm(30),
  MyMediator1 = rnorm(30),
  MyMediator2 = rnorm(30),
  MyOutcome = rnorm(30)
)

# Fit the model
fit <- buzzEBMcontMcontY(
  model = "MyOutcome ~ MyPredictor | MyMediator1 + MyMediator2",
  dataset = my_data
)

Longitudinal Bayesian Mediation

library(buzzMed)

# Load the example longitudinal dataset
data(sublongspikes)

# Fit the longitudinal Bayesian mediation model
# For model specification, slice 1 is the predictor, slices 2--20 are candidate mediators, and slice 21 is the outcome.
# n.burnin and n.iter are optional arguments
results <- longBMed(
  model = "21 ~ 1 | 2:20",
  data = sublongspikes,
  n.burnin = 100,
  n.iter = 500
)

summary(results)

Included Example Datasets

• singlespikes: Cross-sectional mediation dataset containing one predictor, nineteen candidate mediators, and one outcome.
• sublongspikes: Longitudinal mediation dataset stored as a three-dimensional array for repeated-measures analysis with longBMed().
• framing2: A modified version of the framing dataset from the mediation package containing dichotomized candidate mediators for demonstrating the GT-exploratory Bayesian mediation functions.

Citation

If you use buzzMed in your research, please cite:

Shi, D., Shi, D., & Fairchild, A. J. (2023). Variable Selection for Mediators under a Bayesian Mediation Model. Structural Equation Modeling: A Multidisciplinary Journal, 30(6), 887–900. https://doi.org/10.1080/10705511.2022.2164285

License

This project is licensed under the GNU General Public License v3.0.

See the LICENSE file for details.